Human Segmentation(humanSegmentation)

The Human Segmentation node performs semantic human segmentation on an input image using a selection of pretrained models. It produces a segmentation result image, a mask covering the selected parts/components, and a bounding box around the detected region.

Inputs

image

The input image to segment.

method

The segmentation model to use. Available methods:

• selfie_multiclass_256x256
• human_parsing_lip
• human_parts (deeplabv3p)
• segformer_b3_clothes
• segformer_b3_fashion
• face_parsing

Each method has its own label set, and the available mask_components are populated based on the selected method.

confidence

Minimum confidence threshold for segmentation predictions. Default: 0.4
Range: 0.05 – 0.95
Step: 0.01

Lower values include more uncertain regions in the mask; higher values suppress low-confidence predictions and produce tighter, cleaner masks.

crop_multi

Multiplier used to control the crop region around the detected segmentation. Default: 0
Range: 0 – 10
Step: 0.001

0 keeps the crop tightly bound to the mask/bbox. Higher values expand the crop area around the subject.

mask_components

Multi-select list of segmentation components/classes to include in the mask and result image. Multiple components can be selected; their regions are combined into the output mask.

Available options depend on the selected method, because each model exposes different semantic classes.

Outputs

IMAGE output

The segmentation result image — the visual output of the selected segmentation after applying mask_components, confidence, and crop_multi.

MASK

The mask generated from the selected mask_components after the confidence threshold is applied. It corresponds spatially to the output image.

BBOX

The bounding box around the selected segmentation components, in image coordinates. This can be used for further cropping, conditioning, or region-based processing.

Usage Notes

• Choose a method that matches the type of segmentation you need. For example, clothing-focused methods are useful for fashion-related masks, while face_parsing is suited for facial regions.
• Leave crop_multi at 0 for a tight subject crop, or increase it to include more surrounding context in the result image.
• mask_components is a multi-select, so you can combine several body parts or object classes into a single mask.
• The selected method’s model weights may be downloaded on first use, so the initial run may take longer than subsequent runs.